Multi-turn math reasoning on SHARDED
82.4Accuracy+ FiC
Evaluation Results
| Method | Links | |
|---|---|---|
| + FiCModel=PHI-4-14B2026.05 | 82.4 | |
| + FiCModel=QWEN2.5-7B2026.05 | 78.7 | |
| + FiCModel=OLMO-2-13B2026.05 | 77.1 | |
| + FiCModel=LLAMA-3.1-8B2026.05 | 71.7 | |
| + PROMPT-SELFCHECKModel=PHI-4-14B2026.05 | 68.2 | |
| + PROMPT-RULEModel=PHI-4-14B2026.05 | 67.9 | |
| + GATED-SELFModel=PHI-4-14B2026.05 | 63 | |
| + PROMPT-RULEModel=OLMO-2-13B2026.05 | 61.3 | |
| SHARDED (no intervention)Model=PHI-4-14B2026.05 | 61.2 | |
| + FiCModel=LLAMA-3.2-3B2026.05 | 60.4 | |
| SHARDED (no intervention)Model=QWEN2.5-7B2026.05 | 58.9 | |
| + PROMPT-SELFCHECKModel=OLMO-2-13B2026.05 | 56.1 | |
| SHARDED (no intervention)Model=OLMO-2-13B2026.05 | 54.4 | |
| + PROMPT-RULEModel=QWEN2.5-7B2026.05 | 54.1 | |
| + PROMPT-SELFCHECKModel=QWEN2.5-7B2026.05 | 51.6 | |
| + GATED-EXTERNAL (4o-mini)Model=QWEN2.5-7B2026.05 | 45.6 | |
| SHARDED (no intervention)Model=LLAMA-3.1-8B2026.05 | 44.7 | |
| + GATED-EXTERNAL (4o-mini)Model=PHI-4-14B2026.05 | 42.5 | |
| + PROMPT-SELFCHECKModel=LLAMA-3.1-8B2026.05 | 39.9 | |
| SHARDED (no intervention)Model=LLAMA-3.2-3B2026.05 | 39.2 | |
| + GATED-EXTERNAL (4o-mini)Model=OLMO-2-13B2026.05 | 37.9 | |
| + GATED-SELFModel=QWEN2.5-7B2026.05 | 37.7 | |
| + GATED-EXTERNAL (4o-mini)Model=LLAMA-3.1-8B2026.05 | 36.8 | |
| + PROMPT-RULEModel=LLAMA-3.1-8B2026.05 | 31.3 | |
| + GATED-SELFModel=LLAMA-3.2-3B2026.05 | 30.7 | |
| + GATED-SELFModel=OLMO-2-13B2026.05 | 30 | |
| + PROMPT-RULEModel=LLAMA-3.2-3B2026.05 | 29.4 | |
| + GATED-EXTERNAL (4o-mini)Model=LLAMA-3.2-3B2026.05 | 28.5 | |
| + PROMPT-SELFCHECKModel=LLAMA-3.2-3B2026.05 | 27.2 | |
| + GATED-SELFModel=LLAMA-3.1-8B2026.05 | 4.4 |